1
Session 159, Wednesday, February 13, 2019
Sarika Aggarwal, MD, MHCM, Chief Medical Officer, Beth Israel Deaconess Care Organization
Bill Gillis, MS, Chief Information Officer, Beth Israel Deaconess Care Organization
Predictive Analytics for Data-Driven Care Management
2
Sarika Aggarwal, MD, MHCM
Bill Gillis, MS
Have no real or apparent conflicts of interest to report.
Conflict of Interest
3
Learning Objectives
Beth Israel Deaconess Care Organization
Our Vision for Care Management
Our Approach
Predictive Analytics to Identify Impactable Patients
Lessons Learned
Agenda
4
Describe the challenges with a traditional approach to Care
Management
Assess factors that can be used in a predictive algorithm to
determine whether a patient will benefit from a care management
program
Recognize the importance of an underlying high-quality data asset
to a data-driven care management program
Evaluate benefits of standardizing patient assessment and care
planning tools across a large care management staff
Illustrate the importance of clinical-IT partnership at every stage of
a transformational project
Learning Objectives
5
Beth Israel Deaconess Care
Organization
6
Our Mission
Our mission is to move health care forward by engaging providers
in their communities to achieve success in a value-based delivery
system
We are committed to creating innovative, industry-leading best
practices in the clinical, administrative, and financial aspects of
health care.
BIDCO is a value-based physician and hospital network and
Accountable Care Organization (ACO) in Massachusetts.
Beth Israel Deaconess Care Organization
7
BIDCO at a Glance
8
Hospitals
2.2K
Specialists
100
Employees
200K
Covered lives
40
EHR platforms
supported
35.6M
Patient encounters
500
PCPs
$1.5B
Value-based revenue
8
BIDCO Services and Benefits
Population HealthPopulation Health
Performance
Improvement
Performance
Improvement
Analytics and
Reporting
Analytics and
Reporting
Contracting +
Network Management
Contracting +
Network Management
EnrollmentEnrollment
Hybrid model for community-based, high-risk complex care
management and disease management; homecare and SNF Quality
Collaborative for care coordination; ED utilization for avoidable admits
Hybrid model for community-based, high-risk complex care
management and disease management; homecare and SNF Quality
Collaborative for care coordination; ED utilization for avoidable admits
Performance Improvement Facilitator for each practice; EHR
Optimization program; practice redesign facilitation for quality, TME
programs; facilitate understanding of medical economics dashboards
Performance Improvement Facilitator for each practice; EHR
Optimization program; practice redesign facilitation for quality, TME
programs; facilitate understanding of medical economics dashboards
Quarterly financial performance reporting for Risk Units; desktop
access to population health management tools; medical economics
dashboards; data surveillance
Quarterly financial performance reporting for Risk Units; desktop
access to population health management tools; medical economics
dashboards; data surveillance
Provide exceptional customer service to our network; negotiate risk and
non-risk contracts; negotiate reinsurance contracts; liaison between
BIDCO members and health plans
Provide exceptional customer service to our network; negotiate risk and
non-risk contracts; negotiate reinsurance contracts; liaison between
BIDCO members and health plans
Initial enrollment in BIDCO with health plans; recredentialing;
enrollment data management between BIDCO systems, and between
BIDCO and other entities (HPC, GIC, etc.)
Initial enrollment in BIDCO with health plans; recredentialing;
enrollment data management between BIDCO systems, and between
BIDCO and other entities (HPC, GIC, etc.)
9
Our Vision for Care
Management
10
Increasing Prevalence of Chronic
Disease
1. About 133 million Americans
45% of the population have at
least one chronic disease
2. Chronic diseases are responsible
for 7 out of 10 deaths in the U.S.
3. Chronic diseases can be disabling
and reduce a person’s quality of
life
4. Chronic disease accounts for 81%
of hospital admissions; 91% of
all prescriptions filled; and 76% of
all physician visits
11
BIDCO Population Health Strategy
Moderate Risk
Disease
Management
Transitions of
care (30 days)
End of life care
Wellness/Prevention
Programs
BH and MH
Chronic Disease
programs
DATA ANALYTICS
Population Health
Management
Goal: Improve care, reduce
costs, and promote wellness.
Focus on entire health care
continuum
Focus interventions
appropriate for each risk
category
Leverage an interdisciplinary
team
Leverage technology to
generate necessary analytic
data and measure performance
5%
15%
80%
High Risk
Complex Case
Management
Low Risk
Primary
Prevention
Target Population
Medicare
Medicare Advantage
Commercial - Self
Insured
Commercial - Fully
Insured
Medicaid
12
Care
management
Care
management
Transitions of
care
Complex Care
Management
Palliative care
management
ED management
Advanced care
planning
Rising risk
management
Rising risk
management
Pharmacy
Collaborative
program
Depression
Collaborative
program
Medication
management
activities including
polypharmacy
Self Management
Action plan
program
Quality &
performance
improvement
Quality &
performance
improvement
Performance
Improvement
facilitation
Risk Unit, Pod &
practice meetings
Patient outreach
and engagement
Payer formulary
management
Medication
adherence
program
Acuity
documentation
Acuity
documentation
Education and
training
Workflow
redesign
Acuity gap tool
training
Medical
Neighborhood
management
Medical
Neighborhood
management
Preferred SNF
and homecare
network
3-day SNF Waiver
Post-acute Self
Management
Action plan
program
Community
partner programs
BIDCO’S POPULATION HEALTH PROGRAMS
13
Care Management at BIDCO
DM: 2
Pharmacists, 4
Coaches
Quality: 8
Performance
Improvement
Facilitators
BH: 2 LICSW, 0.2
FTE Psychiatrist
COPD/CHF: 4 RN
Rising Risk
Delivery Model:
Telephonic, face to
face
Embedded in
Health Center,
PCMH and
hospitals for TOC
TOC: 3 RN, 2
Pharmacists, 1
CRS
Complex CM: 22
RN
Coordination:
CHW, CRS
Complex CM
Delivery model:
Delegated and Non-
delegated;
Embedded and Remote
Structure and Delivery Model
RN-led care management program to facilitate care coordination and medical management for
complex patients
Team: Consists of Social Workers, pharmacists, non-clinical workers
Delivery Model: Delegated/Non-Delegated, Embedded, Telephonic
Risk Stratification: Payer, BIDCO, Provider groups
Documentation for Care Management: Free text notes in EMR, multiple non-standardized
assessments
14
Traditional Care Management Challenges
2
Patient Selection
Traditional morbidity-based risk
algorithms do not incorporate data
needed to select patients likely to
benefit.
3
Inconsistent Approaches
Each care manager has an
individual approach to assessment
and care plan development
meaning best practices may not
be consistently applied.
5
Unmeasured Performance
Limited insight into outcomes of
various approaches to care
management or of individual
care manager performance.
1
Limited Resources
Valuable part of population health
management, but not enough
resources for all rising risk or high
risk patients.
4
Multiple EMRs/Platforms
Multiple EMRs and other systems
mean that there are multiple
areas of documentation for care
management.
6
Throttled Throughput
Non-standardized documentation
limits the efficiency of a care
management operation, reducing
the number of patients it can
serve.
15
Leverage
clinical
data asset
Our Vision for Care Management
Identify impactable patients
Implement consistent, standardized,
evidence-based workflows and care paths
Measure outcomes to identify best
practices
EMR-agnostic: integrate across multiple
EMRs
11
44
22
Match appropriate patient with appropriate
program
33
55
Manage performance (operations/throughput
and care management outcomes)
66
77
16
1. Establish strong clinical-IT partnership
2. Leverage existing investment in technical architecture
3. Work with Care Managers to identify needed data elements
4. Ensure NCQA standard elements for payer delegation
5. Standardized detailed assessments and templated approaches
6. Establish policies, procedures, care paths and self management action
plans for each program
7. Automate routine tasks
8. Build program-wide reporting and performance monitoring
9. Leverage predictive analytics to customize risk stratification for each
distinct program
10. Use risk stratification to align program resources with patient case loads
BIDCO’s Implementation
17
Our Implementation
18
1. Strong Clinical-IT Partnership
CMO-CIO Collaboration
Tight partnership at the leadership level
fosters collaboration between teams.
Day 1 Input
Care Management team
had direct input at the very
start of the IT project.
Weekly Meetings
Regular communications
between clinical and IT team
members throughout the
project.
Job Shadowing
IT team spent days
shadowing care managers
for a first hand look at
challenges and tasks.
19
BIDCO’s Data Foundation
Pioneer in EHR data
aggregation
High quality clinical data
asset trusted by providers
BIDCO Goals for Care
Management
Use clinical data asset to
support better care
management
Better identify actionable
patients
Capture structured care
management data in for
analysis and reporting
2. Leverage Technical Architecture
CENTRALIZED
INTEGRATION
MANAGER
DATA
NORMALIZATION
RULES ENGINE
100+ EHR
SOURCES
BIDMC Hospital
EHR #1 (homegrown)
BIDMC Hospital
EHR #1 (homegrown)
BIDCO Employed
Physicians
EHR #2
BIDCO Employed
Physicians
EHR #2
BIDCO Employed
Physicians
EHR #2
BIDCO Employed
Physicians
EHR #2
BIDCO Employed
Physicians
EHR #2
BIDCO Employed
Physicians
EHR #2
BIDMC Newly Acquired
Hospital
EHR #2
BIDMC Newly Acquired
Hospital
EHR #2
Affiliate CHCs
EHRs #2 & #3
Affiliate CHCs
EHRs #2 & #3
Other Affiliates
EHRs TBD
Other Affiliates
EHRs TBD
Claims Feeds
CMS and Commercial
Claims Feeds
CMS and Commercial
ENTERPRISE
CLINCIAL DATA
ASSET
CLAIMS DATA
(250K+ LIVES)
20
2. Leverage Technical Architecture
Care Managers see a longitudinal patient record with clinical, claims-based, and
care management history from our core analytics platform.
21
3. Identify the Right Data Elements
CLINICAL-IT COLLABORATION
Patient Experience
IT worked with care managers to
identify data elements that would
help them create a more positive
experience for patients
ADT
Labs
Scheduling
CORE
ANALYTICS
PLATFORM
Radiology
Hospital
Inpatient
Transfer/
Discharge
CARE
MANAGEMENT
APPLICATION
22
NCQA certifies payers for care management, but supports standardization for providers.
NCQA’s 8 key elements to support care management are truly clinically impactful.
4. Ensure NCQA Standard Elements for
Payer Delegation
Building a
population
health
infrastructure
Allows taking
delegation
from payers
Avoids
duplication
of effort on
both sides
Reduces the
total cost of
care
Why take delegation?
NCQA: Standard Elements for Care Management
23
5. Standards and Templates
User experience is
guided by
standardized,
consistent,
templated care
plans and workflows
that yield structured
information
24
5. Standards and Templates
Care managers can
customize prebuilt
care plans for
individual patients
25
6. Policies Established for Programs
It’s not just about the technology. BIDCO implemented
and documented an operational infrastructure.
Policies and
Procedures
Workflows and
Care Paths
Self-Management
Action Plans
CM Enrollment &
Assessment
Case Closure
Med Reconciliation
Care Planning
Hospital Transition of Care
ER Transition of Care
Medical Neighborhood
Advance Directives
SNF Waiver Admission
and Transition of Care
Referral to IDT Team
Elder Abuse
Home Safety
Delegation Policy
Asthma
COPD
Hypertension
DM
Heart Failure
Enrollment and Case
Closure Workflow
Admission and Non-
Admission Care Path
Asthma
COPD
Hypertension
DM
Heart Failure
Tobacco Use
Hyperlipidemia
26
7. Automate Routine Tasks
Care managers have an
activity feed of scheduled and
assigned tasks
27
8. Program-Wide Monitoring
Using structured data
for care plans and our
central analytics
platform enables us
to provide detailed
performance reporting
and monitoring
28
9. Predictive Analytics
It’s not about finding out things that
are happening right now.
It’s not about finding out exact
outcomes in the future.
It is about using existing information
to identify patterns and to infer trends
and potential outcomes in the future.
How often are my diabetics
going to the ED?”
“Which diabetics are going to
end up in the ED next year?”
“Which diabetics are likely to
use the ED but could be
steered elsewhere?”
29
Which patients are most likely to
respond to care management?
improvements in condition
reductions in cost and utilization
9. Predictive Analytics
A subset of patients are highly
impactable but we need to be able to
identify them.
?
30
Highest risk/cost patients are not
generally impactable with care
management (cancer, accidents)
Traditional risk algorithms are
designed for risk adjustment more
than population stratification
Traditional risk algorithms do not
include all the data needed to
predict who will benefit from care
management
9. Predictive Analytics
Does not attempt to identify the
sickest or highest cost patients.
Can be used in a variety of contexts
and populations.
Can be used to report on diverse
individuals regardless of background.
Can help clinicians identify clusters of
patients within a population for
inclusion in programs
Traditional, morbidity-
based approach
Predictive analytics-
based approach
Population Stratification
31
9. Predictive Analytics
A
B
INPUTS
Demographics
Morbidity Risk
Condition Types
Utilization (OP/IP)
Census Factors
Care Coordination
Population Flags
OUTPUTS
Cost
Utilization
Outcomes
UNMANAGED
POPULATION
ENROLLED
AND MANAGED
A
B
PROJECTED
IMPACT OF
CARE
MANAGEMENT
Given available demographic, clinical, and historical information,
which patients would benefit most under care management?
32
9. Predictive Analytics
EXPECTED CM VALUE:
Cost
Utilization
Health Outcomes
IMPACT
SCORE
The impact score describes
the relative benefit projected
for this patient from care
management.
PROJECTED IMPACT
OF CARE
MANAGEMENT
Given available demographic, clinical, and historical information,
which patients would benefit most under care management?
…and which patients are
most appropriate for which
care management
programs?
33
9. Predictive Analytics
Census
Block
Group in
Cambridge,
MA
Census
Block
Group in
Cambridge,
MA
Population
% Males
% Females
% Under 18
% 18 - 44
% 45 - 64
% 65+
% High School
% Bachelors
% Graduate Degree
Median Earnings (Real Dollars)
Female Earning Ratio (Median
Female Earnings/Median Male
Earnings)
% Population by Race - Native
Hawaiian or Pacific Islander
% Population by Race - American
Indian or Alaskan Native
% Population by Race - Asian
% Population by Race - Hispanic
% Population by Race - Black
% Population by Race - White Non-
Hispanic
Persons per Housing Unit
% Families w/ Incomes < 100% of
Federal Poverty Level
% Families w/ Incomes < 200% of
Federal Poverty Level
% Adults who are Unemployed
% Households Receiving Public
Assistance
% Households w/ No Car
% Households with Children and a
Single Parent
% People Age 25+ w/o High School
Degree
Census Variable Inputs to Risk Model
34
Sweet Spot
Opportunities
Sweet Spot
Opportunities
9. Predictive Analytics: Enhanced Risk Stratification
D
ATA POINT SCORING
RANGE
Mo
rbidity Score 0-8
C
are Coordination Risk 0-3
C
ancer Counterweight 0, -8
P
olypharmacy 0-3
12
-Month IP Utilization
R
ates
0-3
12
-Month ED Weight 0-3
F
railty Weight 0, 3
NS
S7: Income,
E
mployment, Public
Assi
stance,
T
ransportation,
E
ducation,
H
ousehold
Stru
cture
5
MAX
THEORETICAL 28
35
COMPLEX CASE SCORE
MORBIDITY-BASED RISK SCORE
~99% of Patients
in this Box
1% of Patients
to the right of this line
1% of Patients
to the right of this line
1% of Patients
above this line
1% of Patients
above this line
Morbidity Predictor. The cone indicates
patients whose morbidity is a major driver
of their risk score. Those higher on the Y-
Axis within this cone have been shown to
have better success rates from a Care
Management engagement.
Cancer Patients. Patients have been
given a large counterweight when the
primary contributor to their utilization and
morbidity-based risk is high-impact
cancer.
Rising Risk. May be older, frail patients
with many medications and some recent
ED/IP usage who live in poorer
undeveloped neighborhoods. Have
multiple comorbidities, but not in the top
1% based on those factors.
Self Managed. This quadrant represents
those with serious multiple comorbidities,
but who have avoided acute events, live
in more developed neighborhoods, and
are less likely for disease exacerbation.
9. Predictive Analytics: Enhanced Risk Stratification
36
BIDCO uses risk stratification to align program
resources with patient case loads. Every system
needs to maximize use of limited resources. Use a risk score
based on predictive analytics to assign panels to care managers
without manual chart review.
10. Align Program Resources
PATIENTS EACH CARE
MANAGER CAN MANAGE
FOR 3-6 MONTHS.
CARE MANAGERS
EMPLOYED BY
HEALTHCARE SYSTEM
TOTAL PATIENTS IN A
PROGRAM
ADJUST RISK SCORE THRESHOLD TO PRIORITIZE
PATIENTS FOR EACH PROGRAM.
Example: complex case management
.71
=
Patient Score
Ideal population for program yields too many
patients for available care managers.
.8.9 .6
Phase 1: Prioritize highest-scoring patients.
Sample Patient Distribution: Low
risk/impactability to high risk/impactability.
37
Lessons Learned
Clinical-IT
Partnership
Physicians and nurse
champions should be part of
decision making at every
stage so clinical stakeholders
trust the solution.
Look Beyond Disease-
Based Approaches
Stratify and enroll patients in programs
using a range of information beyond
traditional morbidity-based risk scoring.
Leverage investments in clinical data
assets to feed predictive analytics.
Use Structured
Data
Capturing care management
plans in a structured format
enables analysis especially
when other clinical data is
integrated.
38
Sarika Aggarwal, MD, MHCM
saggarwal@bidmc.harvard.edu
https://www.linkedin.com/in/sarika-
aggarwal-md-mhcm-3b243579/
Questions
Please remember to complete the online session evaluation.
Bill Gillis, MS
bgillis@bidmc.harvard.edu
https://www.linkedin.com/in/bill-gillis-
a59731/